LISTEN: Leveraging Open-Weight LLMs for Real-Time Monitoring of Health Misinformation on Social Media

We present LISTEN (Leveraging Intelligence from Social Networks to Promote Health Equity and Neutralize Negative Content), a platform designed to detect and monitor health-related misinformation on X/Twitter. LISTEN collects weekly social media data on multiple disease topics (e.g., mpox, measles, cancer) and classifies misinformation narratives into four themes—symptoms, transmission, treatment, and other—based on a taxonomy developed in consultation with public health experts. Misinformation detection is performed using few-shot in-context learning with open-weight large language models, validated against a manually annotated sample. LISTEN curated over 21,000 unique posts published between July 2022 and June 2025. Key opinion leaders included journalists, health professionals, and influencers, underscoring the importance of monitoring high-visibility accounts in shaping discourse. The misinformation detection module achieved a macro-F1 score of 0.806 (precision 0.556, recall 0.833, accuracy 0.907) with 17-shot prompting, substantially outperforming zero-shot baselines. Stakeholder feedback emphasized the utility of engagement visualizations, rationale transparency, and the potential for integration with bot detection and counter-messaging features. The LISTEN prototype demonstrates the feasibility of integrating interpretable, open-weight LLMs into real-time public health surveillance. By surfacing emerging misinformation narratives and their amplification pathways, the system enables evidence-informed communication strategies to reduce stigma and misinformation harms.